[Fixed Seed]
End-to-End Synthetic Data Pipeline
A take-home challenge: synthetic data → pure-SQL DuckDB pipeline → Streamlit dashboard → the project's knowledge graph served over MCP.
A technical challenge for a data engineering role: build an analytics stack from scratch with free choice of tools, and document how AI was used.
Solo, with a coding agent; every architecture decision recorded and approved before implementation.
Show a complete, reproducible architecture without real data — and be honest about what the AI did and didn't do.
01 · Deterministic data
Python + Faker with fixed seeds: every run produces the same dataset.
02 · Pure-SQL pipeline
DuckDB, raw → validate → clean → gold. No AI calls at runtime.
03 · Decoupled dashboard
Streamlit with the data layer separated from UI code and tested with pytest.
04 · Project as a graph
A knowledge graph of the repository served by two read-only MCP servers.
05 · AI usage log
A living record of tools, prompts, what was accelerated, what a human reviewed and known limitations.
! brokeSome bugs only appeared in the deployed environment, never locally.
→ changedA human checkpoint after each phase: run the real app and read the deploy logs before moving on.